Contour-Based Motion Estimation.

摘要

This paper introduces a contour-based approach to motion estimation. It is based on first computing motion at image corners, and then propagating the corner motion estimates along the principal contours in the image based on a local 2 1/2D motion assumption. The results of several experiments are presented. The earliest problem that arises in the analysis of time varying images is the detection of moving image elements (edge, regions) and the computation of the image velocity (optic flow) of those elements. A variety of computational schemes have been proposed to solve this problem.